This study explores new ways of tag-based personalized recommendation by relieving the assumption that tags assigned by a user occur independently of each other. The new methods profile users using tag co-occurrence networks, upon which link-based node weighting methods (e.g. PageRank and HITS) are applied to refine the weights of tags. A diffusion process is then performed upon the tag-item bipartite graph to transform the weights of tags into recommendation scores for items. Experiments on three datasets showed improvements of the proposed method over the tag-based collaborative filtering in terms of precision and recall in the datasets with dense user-tag networks and in terms of inter-diversity in all datasets. In addition, the user-tag network is found to be a necessary instrument for the improvements. The findings of this study contribute to more accurate user profiling and personalized recommendations using network theory and have practical implications for tag-based recommendation systems.